The present invention relates to communication channel quality monitor systems, and more particularly, to monitoring systems for indicating the intended values of signals in the communications channel if uncorrupted by noise and for indicating the extent of any such corruption occurring.
Communication channels can be entirely analog extending from the information source to the transmitter, and then along the channel link to the receiver, and finally to the user of the transmitted information. Communication channels can also be digitally based, either totally or partially. A typical situation is to have a digital information source but where the information is sent over an analog channel link to a receiver which reconverts the information to a digital format.
Digital communication channels or communication channels having digital portions along the channel, i.e. digitally based communication channels hereinafter, are designed to have as large a tolerance for noise and other signal degradations as practicable. Such a system can have so large a designed-in tolerance that the system will operate substantially error free even though one or more elements thereof comes to operate in a severely degraded manner.
A major objective of performance monitoring is to detect such element degradation before corresponding errors are introduced, to thereby permit corrections or other expedients to prevent any such errors before they begin to occur in the communications channel. Obviously, the desired information indicating degradation cannot be first obtained through examining the output of a digitally based communications channel because the foregoing monitoring objective requires detecting degradations before errors occur in the channel.
True, there are tests possible which remove digital elements in the communication channel from service long enough to subject them to test sequences to measure the error rates of these elements, and further, there can be use made of error detection and correction codes. There certainly are situations in which these techniques are useful in performance monitoring. However, in circumstances of degradation, they are not adequate for measuring the performance margin between the point of error free operation and the point where some selected error rate occurs because these techniques give no indication that degraded signal circumstances exist until the degradation is sufficient to cause errors in the received signals. Thus, a useful signal degradation monitoring technique should be capable of detecting the degradation before it becomes extensive enough to cause errors in the signal received at the end of the communications channel.
For both analog and digitally based communication channels, the ability to be able to detect signal degradation, before such degradation becomes so extensive as to cause errors in the received messages in either kind of communications channel, is of vital importance. In comparison, however, the ability to detect gradually increasing signal degredations, and so anticipate loss of the communications channel, is far more easily available or acquired for analog based communication channels than it is for digitally based communication channels.
In analog communication channels, such as voice channels, signal degradation due to such causes as channel induced noise and distortion is delivered to the information user at the receiver right along with the desired signal, with the result that these degradations are quite detectable by the user. Further, these signal degradations are detectable by the user at power levels several decades lower than the level at which they make the analog communications channel unusable through rendering substantially unintelligible the voice signal received. That is, in an analog communications channel there is typically a large margin between the point at which noise and distortion signal degradations are first detectable and the point where such degradations become intolerable. Finally, the user of a voice channel can readily estimate the degree of signal degradation by a qualitative estimate of signal intelligibility.
On the other hand, the user at the receiver of a digitally based communications channel is presented with a very different situation. This is because each digital element in the communications channel reshapes the digital pulses received by it so that any signal degradation occurring theretofore, due to channel noise, distortion or the like, is removed before the signal is transmitted to the next element in the digitally based communications channel. Such reshaping of the digital pulses is part of the designed-in tolerance to signal degradation provided in a digitally based communications channel as mentioned above. That is, the error rate is reduced by removing the noise and distortion at each digital element in a digitally based communications channel so that the degradation occurring in elements of the digitally based communications channel prior to that element, and in that element, are not allowed to accumulate as they would in an analog communications channel.
Thus, the combination of the noise and distortion for the entire communications channel may be so large as to produce an intolerable error rate across the digitally based communications channel were there are no intermediate pulse reshaping. Yet, the actual error rate across the digitally based communications channel can be reduced to approximately zero by the described removing of signal degradation, the removing accomplished through regenerating the digital signal at some or all of the digital elements occurring along the digitally based communications channel.
Therefore, so long as the accumulated degradation in each digital element is kept below the critical level for that element, each digital element will run error free, and hence, the digital communications channel as a whole will run error free. On the other hand, if degradation in one, several, or all of these digital elements occurs to the extent so as to be just slightly below the critical level at which errors begin to occur in the particular digital element, there will be no indication of an impending problem in the error free signal being delivered to the user at the receiver.
The result is that the reshaping of the digital pulses in the digitally based communications channels is advantageous in reducing the error rate across the channel. Unadvoidably, however, such reshaping removes, in the signal delivered to the user at the receiver, any indication of channel degradation. Hence, the signals received at the channel receiver by the user provide no indication of degradation until errors are actually occurring in these signals. As a result, the user who has nothing but the error rate in the signal obtained by him at the receiver to observe also has no means of estimating how close signal degradation is to critical levels in the digital elements, the levels at which errors begin to occur, until one or more of these levels has been exceeded thereby introducing such errors.
The inability of the user to detect gradual degradation of the signals delivered to him at the receiver, until they contained errors, would be less objectionable if there were a greater separation between the degradation level at which the error rate becmes just barely measureable and the degradation level at which the error rate becomes intolerable. The smallness of this separation can be seen by first assuming that the source of the signal degradation is additive, uncorrelated Gaussian noise so that the amplitude of the noise will be distributed in accordance with the cumulative normal probability function, a graph of which is shown in FIG. 1. (To reduce the vertical extent of the graph, the ordinate axis values are provided on the curve with every seven orders of magnitude of the curve graphed repeatedly on the same vertical axis.)
From FIG. 1, one can observe that the probability function, P(z&gt;t), for the normally distributed noise amplitude, z, to exceed an arbitrary threshold, t, decreases so rapidly with increasing t that even when using a seven decade semilog scale, the probability function crosses seven decades vertically more than seven times (indicating more than 49 decades) as the amplitude of t changes less than 24 db (1.2 decades). The consequence of this extremely rapid change in P(z&gt;t) with respect to t, is that the bit error rate of a receiver at the end of a digitally based communications link can change very rapidly with respect to small changes in the amplitude of the additive Gaussian noise.
For ordinary pulse amplitude modulated signalling (PAM), one can show that the baud error rate (BER), i.e. the probability of receiving one or more bits incorrectly in a single baud or sample period (for additive, uncorrelated Gaussian noise) is as follows: ##EQU1## where L number of levels per baud.
z normally distributed random variable with mean=0 and variance=1. PA1 S r.m.s. signal power at receiver decision circuit. PA1 N r.m.s. noise power at receiver decision circuit. PA1 P[z&gt; . . . ] the probability plotted in FIG. 1.
Where either of the most common types of partial response signalling are used, Class I with n=2 or Class IV with n=3, one can show that the BER for additive, uncorrelated Gaussian noise is as follows: ##EQU2## where ##EQU3## In each of these equations, the receiver is assumed to receive an analog signal which the decision circuit therein determines to be at one of the discrete analog levels that the received analog signal is intended to equal at some point in the sample period. The symbols in this latter BER equation are the same as those used in the first equation with the exception of M which is defined therebelow.
Now from the latter of the above BER equations, one can show for partial response signalling how the BER can change, for a relatively small change in the signal to noise ratio, from a BER value essentially equal to zero to a BER value so large as to be intolerable. The following table is constructed to display this for a Class IV, three level, partial response signalling system having a baud rate of 12.5 megabauds/sec:
______________________________________ Errors/Time BER (S/N) db ______________________________________ 10,000 errors/second 8 .times. 10.sup.-4 13.31 100 errors/second 8 .times. 10.sup.-6 15.89 1 error/second 8 .times. 10.sup.-8 17.52 1 error/minute 1.33 .times. 10.sup.-9 18.60 1 error/hour 2.22 .times. 10.sup.-11 19.46 1 error/day 9.26 .times. 10.sup.-13 20.04 1 error/year 2.54 .times. 10.sup.-15 20.94 1 error/century 2.54 .times. 10.sup.-17 21.53 ______________________________________
The first column of this table presents errors as a function of time which are converted into the baud error rate, BER, in the center column. With the corresponding BER, the signal-to-noise ratio (S/N) at the receiver decision circuit is calculated as shown in the right-hand column of the above table. As can been seen in this table, the difference in S/N required to go from 100 errors/sec. to 1 error/century is only 5.64 db.
Further problems arise for the user at the receiver of the digitally based communications channel should he attempt to rely on detecting degradation in the signal received by him through the channel by the method of observing the errors in this signal as the means of monitoring channel performance. These problems arise because of the number of errors which the user must observe for any meaningful conclusions as to the error rate being experienced.
To obtain a reasonably accurate performance measurement, the user must observe a significant number of errors because the standard deviation of the number of errors measured per observation sample essentially equals the square root of the average number of errors measured per observation sample. By the way of example, if the average number of errors per observation sample is 100, then the standard deviation for this ovservation sample is computed as (100).sup.1/2 =10. This means that the BER is being measured with an r.m.s. error of about 10%, that is, one standard deviation equals about 10%.
Now for observing the errors in the signals delivered at the receiver for an observation sampling period of an hour, the percentage error in the calculated error rate determined from the hour observation will increase rapidly as the error rate in the signal delivered to the receiver drops below 1 error/min as can be seen from the foregoing statements concerning the standard deviation of the observed errors. Also, for observation periods of an hour during which there is sampling of the number of errors occurring in the signals delivered to the receiver, the user will be computing an error rate that is based on error observations which on the average are already half an hour old at the time the computation is made. Yet the S/N producing 1 error/min is only 2.71 db lower than the S/N producing 100 errors/second. This is a very small performance margin between acceptable and intolerable system performance, a margin which may be reduced by the errors in measuring the actual error rate, and a margin which can quickly be overcome by changes in the communications channel that can take place in relatively short periods of time.
A monitor system for monitoring the performance of a digitally based communications channel based on using these error counting methods on the signals delivered at the system receiver, having such a narrow S/N margin between acceptable and intolerable performance, is not a very satisfactory system for a monitor that is intended to predict rather than confirm failure in the channel. And, of course, if a larger observation time is used to increase the error sample for the purpose of reducing the error occurring in the measuring of the error rate, the longer time causes an even longer delay in the monitoring process. This makes it difficult or impossible for the monitor system to keep up with what is presently occurring in the communications channel.
From the foregoing, one concludes that counting errors in the signals delivered to the user at the receiver of a digitally based communications channel is likely to be an unsatisfactory monitoring method for monitoring the performance of the communications channel even though such counting techniques are relatively easily implemented in an electronic monitoring system.
Another well known method for monitoring the performance of digitally based communication channels is to display on an oscilloscope the "eye patterns" developed at the inputs to the receiver decision circuits in a channel using an analog link ahead of the receiver decision circuits. This is accomplished by taking the analog signal from the communications channel just before it is submitted to the decision circuits in the receiver and displaying it on the vertical scale of the oscilloscope, with the oscilloscope horizontal scale (time base) synchronized to the baud rate characterizing the delivered signal.
FIG. 2 shows the resulting oscilloscope pattern in such an arrangement for a Class IV, three level, partial response baseband signal assuming no noise is present with the signal. The three levels represent discrete values of signal amplitude, one of which the received partial response signal is intended to equal at the sampling time in each baud. These three levels are shown having the values +2d volts at the upper expected incoming signal level, zero volts at the center expected incoming signal level, and -2d volts at the lower expected incoming signal level. Such a signal format leads to designating the resulting pattern on the oscilloscope as a three level "eye pattern". The eye pattern has two "eye openings" at each sampling point, with each eye opening bracketed on either side by one of the expected signal levels.
The receiver decision circuits effectively sample the baseband signal at each of the sampling times to decide whether an upper, center, or lower expected incoming signal level was intended to be received at that sampling time, the decisions being based on where the signal amplitude is with respect to the receiver decision circuit thresholds. These thresholds are set normally half way between the expected signal levels, i.e. they are set at +d volts for the upper level decision circuit threshold and at -d volts for the lower level decision circuit threshold. Thus, the "eye openings" are more or less centered around one or the other of the decision circuit thresholds.
As noise and distortion degrade the signal delivered at the receiver to the user, the oscilloscope traces of these signals shown in FIG. 2 will no longer all appear to go through one of the three expected incoming signal levels at the sampling times but at least some will pass at various distances above or below these levels causing the eye openings to shrink. That is, as noise and distortion increase, the oscilloscope traces appear to blur and widen about the upper, middle and lower expected incoming signal levels.
When the widening of signal traces around any of the expected incoming signal levels becomes so wide that there is no longer a clear separation between the top and bottom of the eye openings, the decision circuits will begin to misinterpret the signal delivered at the receiver, and so the intended message carried therein, leading to errors. That is, the baseband signal (incoming signal after final demodulation) obtained from the signals delivered at the receiver may be sufficiently perturbed by noise and distortion to have values at the various sampling points, or times, other than a value of one of the expected incoming signal levels. The deviation from the intended expected incoming signal level may become so great as to pass on the wrong side of a decision circuit threshold value of either +d or -d volts. Then an error will be made by the receiver decision circuit through its assigning the signal at a particular sampling point to an expected incoming signal level other than the intended expected incoming signal level.
The size of the eye openings relative to the distances between the centers of the adjacent expected incoming signal levels, when expressed as a "percentage of eye opening", has long been used as a figure of merit for performance measurement of digitally based communications channels. This is quite a useful performance measurement, but it has limitations. First, if the decision threshold levels are not located in the center of the eye openings vertically in FIG. 2 and, second, if the sampling times are not centered horizontally in the eye openings in FIG. 2, then the receiver will begin to make errors before the eye openings are totally closed. third, since the noise encountered in the communications channel typically has a Gaussian amplitude distribution, the widths of the delivered baseband signals about the expected incoming signal levels (and hence the percentage of the eye openings) is not sharply defined. This lack of definition is because the width of the delivered baseband signals about an expected incoming signal level on the oscilloscope can be varied considerably depending upon the intensity setting of the oscilloscope and the length of the time exposure. Finally, the method of observing an oscilloscope is hardly very easily implemented directly in an electronic system.
Another method for assessing signal quality has been to count the number of sample values over a number of baud periods between two fixed thresholds, such as d and another higher threshold set at 2d-b in FIG. 2, and then taking this count divided by the number of baud periods to be a "pseudo error rate". However, this "pseudo error rate" is generally not a linear function of the signal degradation and no one pair of fixed threshold values seems to give adequate sensitivity for the present purposes for the entire range of degradation encountered. A system more or less along this line for a two level eye pattern is described in U.S. Pat. No. 3,721,959 to George.
What is really desired for a digitally based communications channel monitoring system is to measure that probability density function for the signal perturbations, i.e. deviations, from the expected incoming signal levels (those various levels of baseband signal amplitude which would occur at the sampling points in the absence of degradation) so that the desired error rates and performance margins can be predicted. In actual practice, however, point by point determination of this probability density function is usually economically not feasible.
A practical alternative is to assume that the distribution of perturbations, or deviation amplitudes, from the expected incoming signal levels, due to signal degradation, are Gaussian and to make some measurement with respect to the signal delivered to the receiver from which the rms amplitude of the distribution may be inferred. However, there are several common signal degradation conditions such as additive tones, highly correlated intersymbol interference, and impulse noise for which the distribution of the perturbations will deviate significantly from a Gaussian distribution. Thus, there is a desire to augment the first measurement for inferring rms amplitude with a second measurement which can indicate either that the distribution is Gaussian or indicate the nature of its deviation from being Gaussian.